【作者】 刘军伟；

【导师】 冯焕清；

【作者基本信息】 中国科学技术大学 ， 生物医学工程， 2009， 博士

【摘要】 过去几十年,医学成像技术飞速发展,开展以此为基础的计算机辅助诊断系统研究成为该领域的热点课题,而医学图像分割是计算机辅助诊断系统设计中的关键技术。对于医学图像分割问题,它一般可分为两部分:(1)图像中特定目标区域(器官或组织)的识别;(2)目标区域完整性的描述与提取。相比于其他图像,医学图像的复杂性和多样性,使得传统的基于底层图像信息的分割方法很难取得好的分割结果,图像分割成为计算机辅助诊断系统设计中的瓶颈。而结合了高层视觉先验知识和图像底层信息的主动轮廓模型,符合人类视觉认知事物的一般规律,在计算机辅助诊断中得到了广泛的应用。水平集方法的出现,极大推动了主动轮廓模型的发展。水平集方法和曲线演化模型相结合,克服了传统Snakes模型的很多固有缺陷,大大拓展了主动轮廓模型的应用领域。本文将主要研究如何完善水平集函数演化理论和分割方法,并将该方法推广应用到医学图像分割领域。本文主要的研究内容及其创新点包括:1)针对主动轮廓模型的局限性,本文在对包括Balloon模型和GVF模型进行深入研究的基础上,提出了一种改进的方法,该模型能够适应边缘断裂图像和深度凹陷图像的分割,并且具有收敛速度快等优点。2)针对传统单相水平集函数的几何主动轮廓模型在分割较复杂图像时的缺陷,引入了一种多相水平集函数的分割模型框架,对多目标同时分割方法的研究具有重要的指导意义。提出了一种新的基于边缘信息的全局最优的几何主动轮廓模型模型,通过对比人工图像和肝脏CT图像的分割实验,证明了该方法的有效性和实用性。3)通过推导一般通用模型的分割框架,指出了经典的Chan-Vese分割模型的物理意义。针对肺CT图像分割中存在的难题,在统一的Bayes统计模型框架下,提出了一种新的联合主动轮廓分割模型,成功地对肺部CT图像进行了完整分割,为基于肺部CT图像的肺功能测定、肺配准等研究打下了基础。4)分析医学图像分割的特点,发现这些特点恰恰符合灰度连接理论的“相似性、连接性”特征。为此,本文提出一种新型的图像力场构建方法,将模糊连接度理论引入医学图像的水平集分割框架。通过对信噪比非常低的医学超声图像进行分割实验,证明了本方法的优越性。5)针对胸部CT图像肺内树状细纹理提取难题,本文引入一种新颖的基于局部区域信息的水平集分割方法—DPLBF模型。通过对比实验发现,新方法比其他基于全局区域信息或梯度边缘的水平集方法具有更加准确而完整的分割结果。最后,对一组三维高分辨率CT数据集进行了分割实验,通过三维重建发现,得到的肺纹理树具有良好的各向同一性和完整性。更多还原

【Abstract】 In recent years,medical imaging technique achieves a large development,on which computer aid-diagnosis system(CAD) research is becoming a hot research point.Medical image segmentation is one of most important technique for CAD system.Image segmentation generally contains two item questions.One is the given tissue recognition of body reigion from medical image,and the other is method of description and extraction of object region.In most occation,medical image segmentation is more difficult than other image segmentation,because of the complexity and diversity of the medical imaging techniques.Image segmentation technique is generally considered as the bottleneck of CAD system.Active contours model has been widely used in medical image segmentation,as it is based on human vision theory and image information.Level set method coupled with curve evolution theory conquers many limitations of general active contours model,which widens the utilization of active contours model.This dissertation analyses the latest level set method,and its major contents think much of level set evolving theory and medical segmentation method.The main research work and contribution of this dissertation can be summarized as follows:1) The general active contours model has many limitations,an improved method based on Balloon model and Gradient Vector Flow model is introduced to conquer those limitations,which can segment images with sunken region at a fast convergence speed.2) To realize drawbacks of genearl geometric active contours model,a novel multiphase level set function segmentation method is proposed.As level set methods based on image edge information sometimes get into local optimization, a novel global optimization edge-based level set method is introduced into liver CT image segmentation.3) A new level set segmentation framework is researched and explains Chan-Vese model in a new way.For the lung segmentation problem from high resolution CT images,a novel united active contours model is proposed by a Bayes-statistical approach.The method is found to be much more efficient in lung segmentation than other methods that are only based on boundary or region.4) Medical image segmentation particularity is found to be consistent with fuzzy connectedness segmentation theory in this dissertation,so a new driven forece of geometric active contours model is introduced into medical segmentation.The experimental results of medical image segmentation prove the validity of this method.The correlative contents provide a new method for active contours model research and medical image segmentation.5) High resolution CT slice images of chest contain amount of texture information, which provide powerful datasets for research of computer aid-diagnosis system. But the extraction of lung tissue textures is a challenge task.A novel method based on level set is proposed for the extraction of lung tissue tree textures.The method is fully automatic and effectual.A number of contrast experiments are performed,and the results of 3d surface reconstruction shows the efficacies and advantages of our method for the segmentation of fine lung tree texture structures.At last,the author appreciates National Natural Science Foundation of China (60771007) and Natural Science Foundation of Anhui Education Department (2006KJ097A).更多还原